Google embeds Gemini architecture directly into silicon: Frozen v2 chip announced
Google has officially confirmed the development of a new-generation specialized processor — Frozen v2. This chip will be part of a strategy for hardware optimization of AI workloads, and its key innovation lies in directly embedding elements of the Gemini model architecture into the silicon structure.
This approach radically reduces the number of computational operations and the volume of data transfers required to process user requests. Instead of traditional data movement between memory and compute units, the chip will execute part of the logic directly at the hardware level.
Complement to TPU, Not a Replacement
It is important to emphasize: Frozen v2 is not intended to replace the existing line of tensor processing units (TPUs). The project is conceived as a highly specialized addition aimed at alleviating the shortage of computing power in Google Cloud. As I have repeatedly noted in my analyses, it is precisely the lack of GPUs and TPUs that has forced the cloud giant to turn down profitable contracts with external clients.
According to Google engineers' estimates, the new chip will be able to process six to ten times more tokens per unit of energy consumed compared to the company's latest AI accelerators. This is a colossal leap in efficiency that could fundamentally change the economics of running large language models.
Experimental Status and Timeline
Currently, Frozen v2 is considered an experimental project within the company. Mass production in volumes comparable to universal TPUs is not yet planned. However, the tech giant has set a tentative target date for the chip's deployment — 2028.
There is also an important limitation: the chip will only be able to work with future versions of Gemini if the model's basic architecture remains unchanged. Any fundamental changes to the architecture could render Frozen v2 incompatible.
Market reaction was immediate: on July 20, Alphabet (GOOG) shares rose by 1.5% on the news of the development.
Context: Competitive Pressure and Talent Loss
In recent months, Google's AI division has been under serious pressure. The launch of Gemini 3.5 Pro is delayed, and the company has lost four leading researchers who moved to competitors — Anthropic and OpenAI. Against this backdrop, Chinese AI models are actively strengthening their positions: according to my data, they already account for up to 46% of tokens processed by American companies.
My analysis: The development of Frozen v2 is not just a technical upgrade, but a strategic response to the performance crisis. Google understands that further scaling of models is hitting the physical limits of chips. Embedding the architecture directly into silicon is the only way to maintain leadership, but the project's success will depend on whether the company can retain key specialists and prevent further falling behind Chinese vendors.